Amazon SageMaker AI now supports custom sequence lengths for serverless model customization jobs

Amazon SageMaker AI Updates

Amazon SageMaker AI now allows you to configure the sequence length for serverless model customization jobs. This feature supports fine-tuning for various open-weight models, including gpt-oss, Gemma, Llama, Nemotron, and Qwen.

Previously, sequence length was fixed and unmodifiable. Now, you can set sequence length parameters in the customization job's hyperparameters, up to the maximum limit for your chosen model and technique.

• SFT and DPO jobs: Set the dataset_max_len hyperparameter.
• RFT jobs: Set the max_prompt_length and max_response_length hyperparameters.

This update provides more flexibility to match your training data characteristics, with maximums up to 16K tokens for many models.

What to do

• Navigate to the Models page in Amazon SageMaker Studio to launch a customization job.
• Use the SageMaker Python SDK or Agent skills with your coding assistant.
• Refer to the SageMaker AI documentation for more details.

Source: AWS release notes

If you need further guidance on AWS, our experts are available at AWS@westloop.io. You may also reach us by submitting the Contact Us form.